A quick valgrind run for
from sage.all import sqrt
T2 = sqrt(2)
for b in range(num := 100_000):
C = sqrt(T2)
confirms that it is in pynac:
==3947957== 799,912 bytes in 99,989 blocks are definitely lost in loss
record 1,299 of 1,300
==3947957== at 0x484182F: malloc (vg_replace_malloc.c:431)
==3947957== by 0x1A6D3B07: sig_malloc (memory.c:1898)
==3947957== by 0x1A6D3B07: __pyx_f_4sage_3ext_6memory_sage_sig_malloc
(memory.c:1517)
==3947957== by 0x13D0EC2B: ???
==3947957== by 0x13D0FEFD: ???
==3947957== by 0x1FBA2393:
GiNaC::numeric::integer_rational_power(GiNaC::numeric&, GiNaC::numeric
const&, GiNaC::numeric const&) (numeric.cpp:1621)
==3947957== by 0x1FBA266D:
GiNaC::numeric::integer_rational_power(GiNaC::numeric&, GiNaC::numeric
const&, GiNaC::numeric const&) (numeric.cpp:1614)
==3947957== by 0x1FBA94DC: GiNaC::rational_power_parts(GiNaC::numeric
const&, GiNaC::numeric const&, GiNaC::numeric&, GiNaC::numeric&, bool&)
(numeric.cpp:1692)
==3947957== by 0x1FBAA2AF: GiNaC::numeric::power(GiNaC::numeric const&)
const (numeric.cpp:1916)
==3947957== by 0x1FBBA918: GiNaC::power::eval(int) const (power.cpp:536)
==3947957== by 0x1FB09107: GiNaC::ex::construct_from_basic(GiNaC::basic
const&) (ex.cpp:923)
==3947957== by 0x1FBB9D89: ex (ex.h:314)
==3947957== by 0x1FBB9D89: GiNaC::power::eval(int) const (power.cpp:507)
==3947957== by 0x1FB09107: GiNaC::ex::construct_from_basic(GiNaC::basic
const&) (ex.cpp:923)
==3947957==
On Wednesday, July 5, 2023 at 5:24:27 PM UTC+2 Gonzalo Tornaria wrote:
> This slowly and inexorably goes on. Computing `sqrt(T2)` leaks 32 bytes
> each and every time (asymptotically).
>
> Found by a student who, through no fault of himself, brought down our
> server (unable to ssh in until the OOM triggered -- but since the leak is
> slow it takes a while to trash 16G of swap).
>
> ===
> $ cat memleak.py
> from sage.all import sqrt
> T2 = sqrt(2)
> import psutil
> ps = psutil.Process()
> base = ps.memory_info().rss
> for a in range(1, 10):
> for b in range(num := 100_000):
> C = sqrt(T2)
> mem = ps.memory_info().rss - base
> print(f"{mem/1e6 :.2f} MB ({mem/a/num :.2f} bytes/iter)")
> $ sage memleak.py
> 2.70 MB (27.03 bytes/iter)
> 5.95 MB (29.74 bytes/iter)
> 9.19 MB (30.64 bytes/iter)
> 12.44 MB (31.09 bytes/iter)
> 15.41 MB (30.82 bytes/iter)
> 18.65 MB (31.09 bytes/iter)
> 21.90 MB (31.28 bytes/iter)
> 25.14 MB (31.43 bytes/iter)
> 28.39 MB (31.54 bytes/iter)
> ===
>
> Replace the 10 in the outer loop by something larger at your own peril
> (each outer iteration will take 3.2M so 10_000 should kill a laptop in an
> hour or two).
>
> This is with system sagemath 10.0 but it also happens with 9.6, 9.7, 9.8
> and 10.0 in cocalc.com.
>
> Best,
> Gonzalo
>
>
--
You received this message because you are subscribed to the Google Groups
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to sage-devel+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/sage-devel/eb17a08e-1576-4ab6-b09a-8333e3d55caan%40googlegroups.com.
